基于多尺度上下文信息的遮挡行人检测  

Occluded Pedestrian Detection Based on Multi-Scale Context Information

作　　者：赵世阳 王晓峰[1] ZHAO Shiyang;WANG Xiaofeng(College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《计算机工程与应用》2022年第11期141-149,共9页Computer Engineering and Applications

基　　金：国家自然科学基金(61872231)。

摘　　要：在遮挡场景下的行人检测一直是计算机视觉中的一个棘手问题,由于被遮挡的行人尺度差异大,可见率低,通常会给检测带来极大的挑战。针对这一问题,提出了一种针对行人遮挡检测的模型结构,对基于anchor-free的行人检测方法进行改进。设计了一种提取多尺度上下文信息的结构,通过级联多个不同扩张率的卷积层,使用密集连接实现多尺度特征共享,提取各个区域的上下文信息来解决遮挡问题。此外,为了提高特征的可分辨性,使用通道注意力机制对多尺度特征融合进行自适应的调整。实验结果表明,该方法在Caltech行人数据集的遮挡子集上实现了41.73%的MR^(−2),性能优于其他检测算法。Pedestrian detection in occluded scenes has always been a thorny problem in computer vision.In this case,due to the large difference in scale of occluded pedestrians and low visibility,it usually brings great challenges to detection.To solve this problem,this paper proposes a model structure for pedestrian occlusion detection,which improves the pedestrian detection method based on anchor-free.First,a structure for extracting multi-scale context information is designed.By cascading multiple convolutional layers with different dilation rates,using dense connections to achieve multi-scale feature sharing,the context information of each region is extracted to solve the occlusion problem.In addition,in order to improve the robustness of features,the multi-scale feature fusion is adaptive adjusted using the channel attention mechanism.Experimental results show that this method achieves 41.73%of MR^(−2) on the occlusion subset of Caltech pedestrian dataset,which is better than other contrast detectors.

关 键 词：行人检测 多尺度上下文 通道注意力 anchor-free 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]